Weekly Update for April 2
Sample Slide
- High
- profile
- departures:
- 24
Progress on uploading files to NIST
- for bullet lands NIST wants to have files smaller than 5 MB
- at 20x magnification we are at 5.5 MB in x3p format
- therefore we down sample to a resolution of 1.29 microns per pixel
- next step: upload files to NIST
- problem: upload is one land at a time
Progress on DFSC
- protocol for individual cartridge case capture complete
- protocol for automated multi-cartridge case capture in the testing phase
- first set of cartridge cases scanned
- using private git LFS (large file storage) repository to (a) keep track of file versions and (b) exchange files with CMU
Literature Search
- 3D shoeprints, photogrammetry, 3D scanning tech
- I have killed approximately 1/20th of a tree in the last week.
Updates
- Draft of paper for intervals for random forest + extensions coming soon.
- Handwriting generation program on GitHub organization page.
REU
All 6 students we made offers to accepted! Here’s some info on them:
- All women: 1 ISU, 3 Albany State, 2 Fayetteville State
- Mean GPA is 3.53
- 5 of 6 wrote in their cover letter about interest in graduate degrees
- Majors: FS w/Bio. conc.; FS; Chem., minor in crim. just.; FS & Chem. double major; FS, minors in chem., crim. just.; FS, minor in Chem.
Application stats for comparison:
- 6 from ASU, 6 from FSU, 3 from ISU, 1 from Univ. Iowa.
- Mean GPA is 3.42 and median 3.42
- 13 female, 1 male.
Glass data
- Can discriminate between manufacturers with only 1 element.
- Is this element always different between manufacturers?
- Potential for easy ID of manufacturer of glass.
- Will start writing up what I learned this week.
Updates
Outreach
- Working on rounding out the first chapter of the book!
- To editor by early next week
Groove Identification - Starting a write-up of methods used already
- Linking currently running code with GitHub!
- Identifying characteristics that predict whether there are grooves on each side - Iteratively trimming based on that indicator - Compare robust linear model to rollapply based on trimmed values
- Submitting abstract to present at AFTE - thinking poster!
Updates
- Working on updating current hierarchical model to add clusters:
Updates
- Keep working on NIJ proposal with Martin, Soyoung and Alicia
- Adapt general renamer program in python by Jason to a very specific type of files
- Run another script that checks if the all the names are correct
Renaming
- Happy undergrad renaming files
Renaming
- Happy undergrad renaming files
Updates
Chumbley Score
- Running Simulations for all window sizes for Profiles New level of Coarseness 0.3 (default suggestion 0.25)
- Reason :: Profiles seem to be giving slightly lower error rates compared to signatures.
- Working on the write up.
Updates
- Working on two papers on shoe print matching algorithm
- In addition to corners, also working on features called SURF ( Speeded-up Robust Features).
Updates
SLR = dissimilarity + rarity
- Is there a generally applicable way of incorporating a measure of trace rarity into a SLR?
- Consider a bivariate score: (dissimilarity score, trace rarity)
- What happens in the case where data is univariate and normal?
Simple SLR/Bivariate SLR/LR comparison
Updates
-We do about 1500 path searches per document.
-Takes about 60 seconds per document - ish
-Working on 2 different paths to speed increase:
1. Being careful about how many searches we have use.
2. Limit area we have to search in (change order of processing).